Perspective: Size selected clusters for catalysis and electrochemistry Free

Clusters containing a well-defined number of atoms offer an ideal platform to study catalysis at the atomic level and can provide detailed fundamental insight into catalytic processes that may be hampered by the complexity of catalysts prepared by more conventional methods which typically yield particles with finite distributions in size and composition. Such small uniform particles can also be realistically treated using accurate computational methods such as density functional theory (DFT), not only to achieve a molecular-level understanding of their function but also to design new catalytic materials in an atom-by-atom fashion. Exploiting the advances in synthesis techniques, one can gain control over the size and composition of the catalysts at the atomic level. The size-selected clusters described in this paper are typically generated in the gas phase and, after mass selection, deposited on suitable support materials at well-defined surface coverage (i.e., metal loading),^1,2 an approach that transitions earlier fundamental gas-phase investigations of clusters³ toward the field of heterogeneous catalysis research. These clusters can serve as models of the catalytic active site, containing a large fraction of under-coordinated atoms. The size- and composition-dependent properties of the clusters are also strongly modified by their interaction with the support. As an example, we can consider Cu clusters for which the size dependency in cluster oxidation⁴ and their catalytic properties toward CO₂ conversion⁵ undergo a trend-reversal between gas phase compared to when deposited on supports. The support material and its morphology can have a significant influence on the character and performance of a cluster via charge transfer as well as affecting cluster stability.^6–10 Moreover, alloy clusters can be created, in which composition is manipulated on the atomic scale, creating compositions that are not found as stable species in bulk phases. Minor changes in size and composition such as the addition or removal of a single atom can have a substantial influence on the activity and selectivity of a reaction. Understanding catalytic processes on an atomic level can potentially lead to the development of more efficient and cost-effective processes. There have been multiple instances where the highly active cluster catalysts are able to process reactions in an environmentally benign condition under low temperature and pressure, compared to bulk and nanosized catalysts of identical materials.² Last but not least, individual clusters which contain only a handful of atoms have a high surface-to-volume ratio which, from an applications perspective, can offer an excellent and economic utilization of the metal loading.

Theoretical and computational methods have played and promise to continue playing an important role in this field.^11,12 In fact, not only do they offer a parallel advanced form of characterization that can be contrasted with and usefully complements experiment, but they are also uniquely able to explore high-energy intermediates or transient states which can be important in the catalytic pathway but could otherwise be ignored as they escape experimental characterization. Indeed, the dependence of catalytic activity of size-selected clusters on cluster size, support, and (alloy) composition have all been investigated and rationalized by simulations, and the detailed catalytic mechanisms unveiled in a number of examples.^11,13 In addition, systematic computational searches have also singled out the role played by the shell of adsorbates (reactants, intermediates, products), formed on the initially under-coordinated clusters under realistic reaction conditions, as a decisive factor in determining stability and catalytic properties of these subnanometer catalysts.^13,14

This review is not intended to be, and cannot be, exhaustive, within the allocated space, of all the reactions that have been studied in the literature. In this perspective, we thus limit ourselves to examples for a selected range of reactions which utilized size-selected clusters and provided fundamental insights into the reaction mechanisms of CO oxidation,^15–18 CO₂ conversion,^5,19 dehydrogenation,^6,20 hydrogenation,^21,22 epoxidation,^23–25 electrochemical water splitting,^26–28 and Li–O₂ electrochemistry.²⁹ (For the interested reader, there is a number of other reactions on size-selected clusters, such as the selective cleavage of the C–O–C bond,³⁰ or the size-dependent oxidation and reduction of clusters,⁴ or acetylene trimerization.³¹) As demonstrated in the showcased studies, the synergy between experiment and theory is instrumental to explain some of the crucial questions in the field of catalysis research, such as the dependence of performance on particle size and composition, as well as particle/support interactions. Finally, we will provide a discussion on the scale up techniques to achieve a crossover from these model lab-scale catalysts to industrial applications and outlook.

Size-selected clusters are typically synthesized by employing physical methods, in high vacuum systems where the clusters ions are generated using a gas-phase cluster ion source, a mass selector, and ion beam optics for collimation and transport of the size-selected cluster ions to the substrate where the cluster ions are deposited. The cluster ion source may consist of a DC (direct current) or RF (radio frequency) magnetron sputtering source for generating clusters using a plasma of Ar gas or Ar mixed with a low dose of reactive gas such as oxygen, methane, or hydrogen sulphide to prepare metallic clusters, or metal oxide, metal carbide, and metal sulfide composite clusters, respectively.^32–34 Here, the sputtered metal atoms are cryogenically cooled by collisions with He gas under pressure inside the aggregation region to generate clusters with a tunable size distribution.^35–40 Through operating dual or multiple magnetron sputtering devices independently in a gas-aggregation cell, it is also possible to create mixed clusters.^41,42 The ions exiting the aggregation region are subsequently transported using an ion guide, followed by ion beam optics for collimation and finally size-separated and deposited. The cluster beam containing ions of a specific size is then deposited on the substrate.³⁸ Typical cluster ion beam currents are between tenths and few nanoamperes, and the clusters are soft-landed on the supports with an impact energy tuned to <2 eV/atom. Other commonly used cluster generation sources include laser vaporization where a laser beam interacts with a rotating metal rod to create a plasma at the point of contact.^17,43,44 The carrier gas within the aggregation chamber helps in cluster generation as well as transporting the atoms and clusters from the plasma to a vacuum region. Clusters of mixed composition can also be created through laser vaporization by seeding a reactive gas, such as oxygen or methane, in the carrier gas to form oxides and carbides. Furthermore, mixed clusters can be formed using binary alloy targets or a dual-rod configuration.^45,46 Electrospray ionization is another technique for introducing clusters into the gas phase and concomitantly ionizing these species.^47,48 A solution of the cluster precursor passes a capillary maintained at a high voltage is electrosprayed into a carrier gas forming a molecular beam of ions through supersonic expansion. The precursors are introduced to the gas phase where the cluster aggregation occurs, with ligands attached and subsequently fragmented to remove ligands according to the precursor and bias placed on the capillary.⁴⁹ 

The sources discussed herein produce a wide distribution of cluster sizes, and a mass separation technique is necessary to select a cluster of well-defined composition. Commonly utilized methods for size selection of charged particles are time of flight mass spectrometers (TOF MS),³⁷ radiofrequency quadrupole mass filters,⁵⁰ and electrostatic quadrupole mass filters.⁵¹ 

Typical supports used for cluster deposition are of model and technological relevance, such as oxide crystals and crystalline oxide films, oxide films prepared by atomic layer deposition (ALD) or carbon-based, for example, amorphous carbon, ultrananocrystalline diamond (UNCD), and highly ordered pyrolytic graphite (HOPG). The level of surface coverage of the support with clusters, typically at a fraction of atomic monolayer (ML) equivalent, is controlled by monitoring the flux of charged clusters landing on the support during deposition using a picoammeter. The known amount of deposited material can be then used in the determination of the activity on per cluster or atom basis. After deposition, the cluster samples are characterized and tested in gaseous or liquid environments, under ultra-high vacuum (UHV) conditions and realistic reaction conditions.

The theoretical approach nearly universally utilized thus far in this field is density functional theory (DFT), in one of its several flavours. DFT in fact realizes an optimal compromise between the accuracy and computational effort. Simulations on systems composed of many tens up to few hundreds atoms are nowadays routinely performed on clusters of workstations or high-performance computing systems, thus enabling a realistic description of the composite aggregate formed by the metal clusters, the surrounding shell of ligands/adsorbates, and a piece of the support, then embedded into appropriate boundary conditions simulating the rest of the substrate.⁵² The choice of the exchange-correlation (xc-)functional most appropriate for the system at hand is decisive for the predictive accuracy of the DFT approach. Importantly, such a choice can be guided and validated using sophisticated and computationally more demanding post-Hartree-Fock methods.^53,54 In general, it can be recalled that the development of more efficient and powerful architectures^55,56 promises a much increased accuracy and reliability, and therefore an even more brilliant future to theoretical and computational research.

Using DFT approaches, local energy minimizations and frequency calculations can be performed first, hence predicting equilibrium geometries and energetics and validating the stability of investigated configurations. The potential energy surface connecting these local minima configurations can then be explored, hence locating the saddle points separating the basins of attraction corresponding to the local minima. Once the network of local minima and connecting saddle points is available, transition-state theory⁵⁷ can then be used to predict reaction rates among local minima, which can be inserted into a kinetic Monte Carlo algorithm⁵⁸ to model the complete catalytic process; see, e.g., Ref. 16 for a recent application of this strategy. Clearly, which configurations to select for investigation represents an open and difficult question. Despite the limited number of atoms directly involved in catalysis, structural sampling is in fact a non-trivial issue, not only because unusual atomic (geometric) arrangements can arise in the metal cluster, or the adsorbates, or the support (especially when considering amorphous systems) but also because the stoichiometry of the system at each given catalytic step is not defined a priori but rather results from steady-state reaction kinetics under the given experimental conditions. This leads to an exponential increase in the degrees of freedom to be explored.⁵⁹ To address this issue, global optimization techniques can be utilized,⁶⁰ i.e., systematic searches in the phase space of geometric configurations and stoichiometry, but to be really effective they should be improved and made more appropriate by combining systematic search with real-time evaluation of reaction rates then used to orient the system evolution according to the given experimental conditions, as, e.g., in the reactive global optimization approach.¹² 

The techniques just described are very effective for dealing with systems exhibiting rugged potential energy surfaces (i.e., high energy barriers). In contrast, when the barriers among minima in the energy landscape are comparable with the thermal energy, it makes less sense to define a set of local minima configurations as basins of attraction, and, as an alternative strategy, ab initio molecular dynamic (AIMD) is a more appropriate technique to describe the dynamics and follow the structural evolution in real time of fluxional systems; see, e.g., Ref. 61 for a recent example. AIMD can also be used in general in combination with various acceleration techniques^62,63 to investigate chemical reactions, or simply as a test of the stability of a given configuration with respect to disaggregation and/or diffusion.¹⁴ 

Size-selected Pt₁₀, Ag₉Pt₂, and Ag₉Pt₃ clusters were deposited on ultrathin amorphous alumina support prepared by ALD, and tested under realistic conditions. The CO oxidation activity of all three cluster species tested was among the highest reported^15,16 with a turnover rate (TOR) between 250 and 300 molecules per second per metal atom at 300 °C, as shown in Figs. 1(a) and 1(b). However, at the highest applied temperatures, some decay in the activity was observed for the Pt₁₀/alumina sample. This decay was explained by STM characterization of Pt₁₀ clusters supported on crystalline alumina [Fig. 1(a)] due to a partial immersion of clusters into the alumina support as observed during the course of the reaction and/or poisoning by oxidized species such as carbonate (CO₃).¹⁵ The catalyst deactivation could be completely curbed by switching to bimetallic Ag/Pt clusters Ag₉Pt₂ and Ag₉Pt₃, as confirmed from combined in situ grazing incidence small angle X-ray scattering/temperature programmed reactivity (GISAXS/TPRx) measurements [Fig. 1(b)].¹⁶ 

The Pt₁₀ system also made it possible, importantly, to clarify the crucial role of amorphous supports in stabilizing ultra-small clusters in heterogeneous catalysis.¹⁵ First, Pt₁₀ clusters were deposited on a regular ultrathin alumina film grown on the NiAl(110) surface, which is a crystalline-line phase widely used as a surface science model of alumina. As a crucial point, this crystalline-like film, when subjected to the CO oxidation reaction under conditions identical to those held on the ALD-alumina support, was shown to convert to an amorphous phase apparently similar and in all analogous to the ALD-prepared film. Second, the Pt₁₀ clusters deposited on the ALD-alumina support did not show any sign of diffusion nor sintering even at elevated temperatures, in stark contrast with the same clusters deposited on a regular titania surface, which underwent reaction-induced cluster ripening and thus fast deactivation.⁶⁴ 

DFT calculations [Fig. 1(c)] reveal the details of the potential energy landscape of bare and CO/O₂-covered Pt₁₀ clusters, predicting a 3D structure for Pt₁₀, an energy barrier of ∼58 kJ mol⁻¹ for CO oxidation, and an intermediate Pt oxidation state, in excellent agreement with the findings from the experimental reactivity and GIXANES (grazing incidence X-ray absorption near edge spectroscopy) characterization. The calculations also reveal a pressure (i.e., reactant coverage) dependent localization of the reaction from a top site to an interfacial one. In the case of Ag/Pt bimetallic clusters, the Pt atoms were found to anchor the mixed clusters to the support and to bind and activate two CO molecules, while Ag binds and activates O₂. In the bimetallic clusters, the proximity of Ag and Pt disfavors poisoning by CO or oxidized species such as carbonate (CO₃). As shown in Fig. 1(d), in the low temperature (RT) regime, the reaction proceeds via sequential co-adsorption of CO and O₂, formation of an OOCO oxametallocycle (OXA) which turns first into a CO₃ carbonate, after which a further adsorbed CO reacts with the carbonate to sequentially evolve two CO₂ molecules. At high temperature, the O₂ adsorption energy is not large enough to overcome the loss of entropic (translational) free energy, and Langmuir-Hinshelwood (LH) steps involving O₂ adsorption are replaced by Eley-Rideal (ER) steps.¹⁶ 

Notable in all Pt₁₀, Ag₉Pt₂, and Ag₉Pt₃ cases is that the reaction pathways proceed on a ligand/cluster/support complex exhibiting multiple adsorbed ligands (high coverage) as the catalytically active species. This feature has been advocated as a general phenomenon in heterogeneous catalysis by subnanometer clusters,^12,13,65 and the catalytic complex thus in situ formed has been proposed as the proper analog of the metal complexes used in homogeneous catalysis.⁶⁶ 

Engineering defects on the substrates could provide a new route for increasing the activity of small deposited clusters. The following measurements have been performed under ultra-high vacuum (UHV) conditions, by investigating size-selected small gold octamers (Au₈) softlanded on a well-characterized crystalline MgO(001) surface prepared with and without oxygen vacancies. The Au₈ cluster deposited on the defect-rich magnesia surface was found active in oxidizing CO into CO₂ at temperatures as low as 140 K. The same cluster bound to an MgO surface without oxygen vacancies (defect-poor) was catalytically inactive in CO combustion, as shown in Fig. 1(e). The experimental results supported by quantum-mechanical ab initio simulations show that the (i) the defect sites serve to anchor the deposited clusters more strongly than the defect-free surface and (ii) the defect sites also alter the charge state of the gold clusters leading to increased activity of the clusters through the promotion of the activation of the adsorbed reactant molecules by the formation of peroxo or superoxo species.¹⁷ It should be added that oxygen vacancies have been claimed to be stabilized at the cluster/support interface and to play an important role in CO oxidation via a Mars-van-Krevelen mechanism.⁶⁷ 

The reactivity of the clusters is known to be strongly dependent on their electronic structure. X-ray photoemission spectroscopy measurements on soft-landed Pd clusters deposited on TiO₂ supports revealed that the Pd 3d binding energy varied non-monotonically with the cluster size and that the changes correlated with strong size variations in the CO oxidation activity, as shown in Fig. 1(f). The shift in the binding energy is a result of combined effects like charge transfer between the cluster and support, quantum confinement effects, and metal-metal bonding within the clusters. Electron transfer to the clusters has substantial effects on their ability to activate O₂, which is presumably the rate limiting step in the CO oxidation reaction. The left-hand axis shows the CO oxidation activity, given as the number of CO₂ product molecules desorbing during temperature-programmed reaction (TPR) from each sample. The low reactivity in the case of samples like Pd₁/TiO₂ and Pd₇/TiO₂ (and in context Pd₂₅/TiO₂) correlated with particularly stable valence structures.¹⁸ 

Several other important contributions toward the study of CO oxidation using deposited Ag,⁶⁸ Au,⁶⁹ Pd,^18,44,70,71 Pt,^72–74 and Rh⁷⁵ single element metallic clusters and also tin oxide clusters.⁷⁶ CO oxidation with bimetallic clusters like Ag–Au⁶⁵ and Co–Pt⁷⁷ has also been reported.

Subnanometer-sized Cu_n clusters (n = 3, 4, and 20) provide a route to designing new catalysts for the efficient conversion of carbon dioxide.⁷⁸ The reduction and catalytic performance of the clusters was found strongly dependent on the cluster size. The temperature at which the initially oxidized clusters reduce decreases with increasing cluster size: Cu₃/Al₂O₃ > Cu₄/Al₂O₃ > Cu₂₀/Al₂O₃.⁵ The onset of methanol production correlates with the reduction of Cu_n clusters (as determined by linear combination analysis of the X-ray absorption spectra), which suggests that the reduced phase of Cu catalysts plays a crucial role in the methanol synthesis from CO₂, as shown in Figs. 2(a) and 2(b).^5,19 The oxidation state of Cu clusters in Fig. 2(b) is obtained by performing a linear combination fitting of the XANES spectra of clusters with the XANES spectra of Cu bulk standards as discussed in detail in Ref. 5. Per copper atom methanol activity varies with the cluster size in the order of Cu₄/Al₂O₃ > Cu₂₀/Al₂O₃ > Cu₃/Al₂O₃. As shown in Fig. 2(a), under low partial pressure of reactants (20 sccm flow of 1% CO₂ and 3% H₂ seeded in helium, at a total pressure of 1.25 atm), the methanol production reached its maximum around 225 °C at a turnover rate of about 4 × 10⁻⁴ methanol molecules per copper atom per second. The performance of this catalyst ranks among the most active for this reaction. By removing a single atom from Cu₄ to Cu₃, the rate of CH₃OH production drops dramatically, in excess of 50%. Above 325 °C, methane formation was also observed. The resulting reaction pathway from the predictive DFT calculations is shown in Fig. 2(c). The calculations predict that supported Cu₄ clusters have a very low activation barrier for conversion of CO₂ to CH₃OH resulting from strong binding of the adsorbates on the under-coordinated atoms of the copper clusters, resulting in an energetically lower-lying reaction pathway and a lower barrier.¹⁹ The higher charge state of Cu in Cu₃/Al₂O₃ compared with Cu₄/Al₂O₃ leads to a weaker binding with intermediates, which is ultimately responsible for its lower per atom activity,⁵ in contrary to the opposite size-dependent trend observed for gas-phase clusters. A similar reversal of trends between gas-phase and alumina-supported clusters has also been found for the oxidizability of free and alumina-supported clusters.⁴ 

Subnanometer Pt clusters deposited on alumina were found to be among the most active to dehydrogenate hydrocarbons; however, like many other catalysts, Pt may deactivate by coking in hydrocarbon-rich environments. Nanoalloying of Pt₇ with boron modified the cluster’s alkene-(deuterated ethylene, C₂D₄) binding affinity and reduced its deactivation. The catalyst was prepared using size-selected deposition of Pt₇ clusters followed with the boration of the clusters by their exposure to diborane (B₂H₆).⁷⁹ Thought Pt₇ was more active than Pt₇B, boration made the cluster more stable as demonstrated by six sequential temperature programmed desorption/reaction (TPD/R) cycles under UHV conditions [Fig. 3(a)]. For Pt₇/Al₂O₃, the amount of both C₂D₄ desorbing and D₂ desorbing decreased substantially between the cycles, and post-reaction XPS showed significant carbon deposition.⁷⁹ For the borated Pt₇/alumina, the poisoning by coking was suppressed, no change was observed in C₂D₄ and D₂ desorption, and no carbon deposits were found by XPS. As shown in the accompanying calculations, boration lowers the binding energy of C₂D₄ such that desorption, rather than dehydrogenation and coking, is favored. CO TPD results shown in Fig. 3(b) revealed that for pure Pt₇, the number of CO binding sites is reduced dramatically after the first C₂D₄ TPD/R cycle and continued to drop during the six consecutive cycles, caused by a combination of coke deposition and cluster sintering/restructuring. The borated clusters also showed a gradual decrease in the number of CO binding sites in the sequential C₂D₄ TPD/R runs, which was most likely caused by the sintering of clusters as no C deposits are detected with XPS. MD calculations revealed smaller binding energies of Pt₇B to the support, which explained agglomeration, as well as a greater structural fluxionality of the borated clusters.²⁰ Boltzmann-weighted ensemble-averages of C-binding energies (E_C) of the isomers performed at 450 and 700 K for Pt₇, Pt₇B, and Pt₈ are shown in Fig. 3(c). Pt₈ was included in these calculations because it possesses prismatic isomers like Pt₇B. The temperature dependence of carbon affinity for Pt₇ and Pt₇B was found to have the opposite trend with temperature. For Pt₇, the ensemble-average E_C increases with temperature; that is, the population evolves to include more isomers with high carbon affinity. In contrast, for Pt₇B, higher energy planar configurations shown in Fig. 3(d) (isomers V and VI), in which B is exposed, rather than anchored to alumina (isomers I–IV), exhibit weaker E_C. As a result, the evolving ensemble for Pt₇B has decreasing carbon affinity and therefore increasing coke resistance as temperature rises. For Pt₈, E_C also decreases with temperature but remains much higher than that for Pt₇B. This illustrates an important point that the size- and composition specific properties of cluster catalysts are necessarily ensemble averages and cannot be described by individual structures, even if they are the global minima.

Dehydrogenation of propane to propylene was highly active and selective on Pt_8–10 clusters stabilized on high surface area supports made of commercial porous anodized aluminum oxide (AAO, Anopore) membranes coated with ALD (atomic layer deposition) alumina with an activity enhancement of up to 40–100 times compared to bulk Pt and vanadia based catalysts, studied under realistic reaction conditions.⁶ Other examples of dehydrogenation reactions include the use of partially oxidized Co clusters in the oxidative dehydrogenation of cyclohexene, interrogated using various supports and reaction conditions.^9,80

Structure sensitivity may play a key role in some reactions of subnanometer clusters. Size-selected Pt_n (n = 8–15) clusters soft-landed on magnesia illustrated this propensity in the hydrogenation of ethylene to ethane under UHV conditions, accompanied by a parallel hydrogenation–dehydrogenation of the ethylidyne-producing route.²² The TPR spectra for ethane production presented in Fig. 4(a) for Pt_n clusters show negligible reactivity of the Pt₉ cluster, while a sizeable ethane production for Pt₁₀ and Pt₁₃ peaking at ∼150 K, that is, ∼100 K below the temperature measured on the extended Pt(111) surface. The results indicate similar activation energies for the clusters active in this size range which are considerably lower than the one found for the extended Pt(111) surface. The integrated areas under the measured TPR curves, normalized to the number of atoms per cluster, are shown in Fig. 4(b). This further illustrates that the supported platinum clusters catalyse the formation of ethane per Pt atom more efficiently than the Pt(111) surface. The Pt₁₃ cluster was found to be the most active catalyst, with activity decreasing monotonically for cluster sizes both larger and smaller than Pt₁₃. The activity seized for clusters smaller than Pt₁₀. Extensive first-principles steered reaction-pathway (SRP) simulations have been done to gain insights into the microscopic ethylene hydrogenation mechanism. For Pt₉, no low-activation-energy hydrogenation channel was found in the simulations in agreement with the TPR results [Fig. 4(a)]. On the other hand, multiple reaction pathways, characterized by relatively low activation barriers, were found in SRP simulations for the larger clusters, as illustrated in Figs. 4(c)–4(e) for Pt₁₃. The simulated pathways starting from a π-bonded ethylene molecule co-adsorbed with a dissociated H₂ on the Pt₁₃ cluster are shown in Fig. 4(c). For both cluster sizes, we observe that two successive energy barriers denoted as ΔE_T^(1,2) correspond to the two hydrogenation steps of the Horiuti-Polanyi (HP) mechanism, whose heights are sufficiently low to permit the experimentally observed low temperature. In addition to the low-activation-energy, pathways found on the Pt₁₃/MgO several reaction sites have been found which are characterized by higher ethylene-hydrogenation activation energy barriers; typically, ΔE_T > 0.8–1.0 eV. On Pt₁₃/MgO, multiple high-energy ethylidyne formation pathways are also shown in Figs. 4(c) and 4(d), manifesting energy-barriers of ΔE_T ∼ 1.2–1.3 eV. This finding implies for Pt₁₃ an onset of significant ethylidyne formation at high temperatures of ∼350–400 K, which can poison the catalyst during ethylene hydrogenation.²² Interestingly, previous theoretical studies of unsupported Pt₃ cluster had investigated the competition between π-bonded adsorption and σ-bonded adsorption of ethylene on small Pt clusters, suggesting that the former adsorption mode is favoured in the “coordinatively saturated” regime, whereas the latter is favoured in the “coordinatively saturated” regime, and that the energy barriers of reaction channels such as oxidative insertion of Pt into C–H bonds with the corresponding formation of platinum hydride species are lower in the latter regime, thus entailing a dependence of the interplay between hydrogenation and dehydrogenation upon ligand coverage.⁸¹ 

Further experiments have been performed to understand the effect of the thickness of the support thin film metal oxides on the morphology and chemical properties of Pt₁₃ clusters. Thin amorphous silica films (a-SiO₂) were grown in UHV to obtain three types of a-SiO₂ films on a metal single crystal: silicon-rich, stoichiometric, and oxygen-rich.⁸² The turnover frequency (TOF) of Pt₁₃ was measured at 300 K (blue) as the function of the a-SiO₂ support stoichiometry using a pulsed molecular beam (PMB) technique,⁸³ as shown in Fig. 4(f). These results are compared with the TOF measured at 300 K after pulsing a defined amount of ethylene at 400 K (green). A comparison is performed also with a Pt(111) single crystal. A distinct reactivity trend was observed before and after the 400 K step, which correlates with the support stoichiometry. At 300 K, the Pt₁₃ clusters on both silicon-rich and stoichiometric films exhibit an activity comparable with the Pt(111) single crystal. However, the activity of the clusters increases three-fold in the oxygen-rich film. The next set of measurements was performed after pulsing a defined amount of ethylene at an elevated temperature of 400 K and the TOFs measured again at 300 K (green). A total deactivation is observed for the clusters supported on the silicon-rich film, while those on the oxygen-rich film lose about half their activity. No deactivation is found for clusters supported on the stoichiometric film, while the Pt(111) single crystal loses some activity. First principles calculations provide insights into the structural and electronic characteristics of Pt₁₃ clusters adsorbed on an a-SiO₂ surface.²¹ The silicon-rich a-SiO₂ film donates electron density into the cluster, thereby filling unoccupied states closer to the Fermi level, and thus lowers the reaction barrier compared to the stoichiometric film. The complete deactivation of the activity at 400 K is caused due to a complete dehydrogenation of ethylene to ethylidyne and graphitic-like species. This is in marked difference with the oxygen rich film which withdraws electrons from the adsorbed clusters, causing a local shift in the energy levels which in turn increases the dehydrogenation activation barriers on Pt₁₃ clusters. The enhanced activity at 300 K [Fig. 4(f)] is unprecedented as the ethylene hydrogenation is a structure insensitive reaction as found on the other two samples. At higher temperatures, the first dehydrogenated products start to form on the oxygen-rich support which results in a significant decrease in the activity observed after pulsing at 400 K.

Size-controlled Pd nanoparticles have also been used for the hydrogenation of furfural as shown in Ref. 84, while the influence of temperature and H₂ adsorption on the structure of silica-supported gold hexamer and heptamer clusters has been investigated in Ref. 85 with results in excellent agreement with experimental findings.

Three-atom silver clusters were softlanded at energies below 1 eV per silver atom on an amorphous ALD-alumina support. The stability of the clusters under reaction conditions was monitored in situ by grazing incidence small angle X-ray scattering (GISAXS) and their catalytic performance by using an in-line mass spectrometer, under atmospheric pressure. The Ag₃ clusters were found to undergo sintering at temperatures exceeding 120 °C [Fig. 5(a)], leading to the formation of stable aggregates (∼2 nm tall, ∼4 nm diameter). The dashed lines in Figs. 5(a)–5(c) indicate the onset temperature of the sintering of the silver trimers in a gas mixture of propylene and oxygen seeded in helium. The silver trimers showed an onset of propylene oxide formation already at temperatures as low as 50 °C, accompanied by the production of acrolein [Fig. 5(b)]. At higher temperatures, combustion to CO₂ prevailed, indicative of a highly efficient low-temperature activation of molecular oxygen on the silver trimer. Density functional theory (DFT) calculations performed on the Ag₃/alumina system revealed the crucial role of the cluster–alumina support interface in the efficient activation of molecular oxygen on the Ag₃ cluster, as well as the role of atop and interfacial oxygen atoms in the epoxidation versus acrolein/CO₂ channel, respectively.²³ Ethylene epoxidation over Ag₃ clusters deposited on a different oxide support (MgO) has also been theoretically investigated, and high-coverage catalytic ligand/cluster/support complexes singled out for this reaction.^12,66

To investigate the origin of high reactivity in Ag aggregates, O₂ dissociation and propylene epoxidation were studied on alumina-supported Ag₁₉.²⁴ The O₂ dissociation and propylene epoxidation apparent barriers at the interface between the Ag aggregate and the alumina support are calculated to be 0.2 and 0.2–0.4 eV, respectively as shown in Figs. 5(d) and 5(e), respectively. These barriers are somewhat lower than those on sites away from the interface. The mechanism at the interface is akin to what was found with silver trimer on alumina and can account for the high activity observed for the direct oxidation of propylene on the Ag aggregates. Due to lower barriers for oxygen dissociation at the interface and on the surfaces of the small Ag particles, the availability of oxygen is not the rate determining state, which is usually the case with crystalline surfaces.

Propene epoxidation has been also studied using Au clusters as catalysts and revealed an important role of hydrogen and water in tuning the activity and selectivity between the propylene oxide and acrolein formation.⁸⁶ 

Co₄ and Co₂₇ clusters supported on a thin iron-oxide film made by ALD on the top of a Si wafer have been used to understand the role of individual catalytic sites in the complex water oxidation reaction.⁸⁷ Electrochemical and electrophotochemical measurements were performed at pH 13.6 to identify the oxygen evolution reaction (OER) activity of the samples. Figures 6(a) and 6(b) show the normalized linear sweep voltammetry (LSV) data taken under dark and illuminated (light) conditions for bare silicon, annealed Si/Fe₂O₃, and annealed Si/Fe₂O₃/Co₂₇ samples. The annealed Si/Fe₂O₃/Co₂₇ sample shows significantly improved electrochemical performance under both dark and illuminated conditions. The photocurrents plotted in Fig. 6(c) show that neither the activity (dark) nor the photoactivity (light) of the thin Fe₂O₃ film changes after the annealing of the Si/Fe₂O₃ sample. The activity of the Si/Fe₂O₃/Co₂₇ sample was found to be three-fold higher than that of the (ALD)-deposited submonolayer Co film on a Fe₂O₃ support.⁸⁸ Also, the onset potential for oxidation under both dark and light remained unchanged for the Fe₂O₃ electrode. In brief, Co₂₇ clusters deposited at very low surface coverage on Fe₂O₃ anodes were significantly more active than Fe₂O₃ alone as a water oxidation catalyst in alkaline media, regardless of whether the anode is used as an electrocatalyst or as a photoelectrocatalyst. Grazing incidence X-ray absorption near-edge spectroscopy (GIXANES) characterization of the anodes before and after electro(photo)chemical cycling demonstrated that these Co₂₇ clusters are stable to dissolution even in the harsh water oxidation electrochemical environment. It is hypothesized that in the annealed cluster sample Si/Fe₂O₃/Co₂₇ the much improved activity may be a result of a better charge transfer between the film and the cluster moiety.²⁶ 

Other cluster based materials studied for electrochemical applications were transition metal dichalcogenides (TMDs). Ni–MoS₂ hybrid nanoclusters were prepared by the dual-target magnetron sputtering and gas condensation process.²⁷ The STEM images in Fig. 6(d) show (top to bottom) the morphologies of MoS₂ nanoclusters, Ni–MoS₂ hybrid nanoclusters, and Ni nanoclusters. A Z-contrast high angle annular dark field - scanning tunneling electron microscopy (HAADF-STEM) imaging has been performed to find the size distribution of the deposited nanoparticles. Figure 6(e) shows the linear sweep voltammograms acquired in the 0 to −1.2 V range [normalized vs normal hydrogen electrode (NHE)] at a scan rate of 25 mV s⁻¹ in a 2 mM HClO₄/0.1M NaClO₄ aqueous electrolyte for all samples tested. The activity of Ni–MoS₂ hybrid nanoclusters is on par with previous reports of electrocatalytic enhancement to hydrogen evolution reaction (HER): an almost 3-fold increase in exchange current densities along with a significant shift in the onset potential (approximately 100 mV), compared to MoS₂ nanoclusters. The Tafel slope shown in Fig. 6(f) is almost unaffected (≈120 mV dec⁻¹). The high activity is achieved when Ni is oxidized after the exposure of the samples to air. This has been further confirmed by XPS measurements.²⁷ 

Studies with electrocatalytic water oxidation with subnanometer Pd⁸⁷ and Pt⁸⁹ clusters have also been studied in detail. The photocatalytic hydrogen evolution reaction with size-selected Pt clusters is presented in Ref. 90. Pt clusters have also been studied for oxygen reduction,^91–94 confirming an extreme dependence upon the precise size of the clusters⁹³ and ethanol electro-oxidation.⁹⁵ 

Li–O₂ batteries are of great interest and provide a new concept of electric energy storage which involves an electrochemical formation of Li–O chemical bonds. These materials are a promising candidate for the long-range electric vehicles; however, its success relies on proper understanding of the charge and discharge chemistries.²⁹ Size-selected silver clusters of four sizes, Ag₃, Ag₉, Ag₁₅, and Ag₁₈, were deposited on alumina-passivated carbon support to study the effect of cluster size on the discharge process in Li–O₂ cells. Figures 7(a) and 7(b) show large and small area transmission electron microscopy (TEM) images of the electrode with deposited Ag₁₅ clusters, showing evenly distributed and intact clusters, respectively. The study revealed dramatically different morphologies of the lithium peroxide (Li₂O₂) grown in cells depending on the size of the applied silver cluster. The resulting different nanoscale to microscale morphologies in the discharge product were understood to be due to the effect of the cluster size on the formation mechanism of lithium peroxide, which also affects the charge process. Out of the four electrodes prepared with Ag₃, Ag₉, Ag₁₅, and Ag₁₈ clusters, the electrode with Ag₁₅ clusters possesses the largest charge capacity [Fig. 7(c)], moreover at the lowest charge potential of ∼3.6 V [Fig. 7(d)]. Although the sizes of all clusters used were below a nanometer, their effect on the morphology of the micrometer-sized discharge products [Figs. 7(e)–7(h)], as well as on the nm-size building blocks of the discharge products [Figs. 7(i)–7(l)], was dramatic. The different morphologies of the Li–O₂ discharge products provide evidence, supported by detailed DFT calculations, for different discharge mechanisms which involve oxygen reduction followed by the formation of either lithium superoxide (LiO₂) or lithium peroxide (Li₂O₂) in the electrolyte, completed by surface nucleation and growth of the final discharge product from the electrolyte. The two discharge mechanisms proposed are either a fast oxygen reduction rate (ORR) as shown in Fig. 7(m) or a slow ORR as shown in Fig. 7(n). These explain the relation between the morphologies of the discharge products and different charge overpotentials with changing cluster size. This study of well-defined clusters on cathode surfaces provides evidence that the nature of the subnanometer surface species can strongly influence the morphology of the discharge products in Li–O₂ cells. These results imply that characterization of the surface structure of cathodes at an atomic/subnanometer level is a key factor in understanding the discharge and charge chemistries in lithium–air cells, and the tailoring of this structure is a potential avenue to improve the performance of Li–O₂ batteries.²⁹ 

Size-controlled gold clusters, here prepared by an alternative solution-based method, supported on functionalized carbon nanotubes demonstrated high catalytic activity toward the oxidation of thiophenol to disulphide in the presence of O₂.⁹⁶ Isolated gold atoms supported on functionalized multi-walled carbon nanotubes (MWCNTs) were synthesized by controlling the pH value of the medium and the concentration of the HAuCl₄ precursor. The aberration-corrected HAADF-STEM image of the sample is shown in Fig. 8(a). The TOF was measured as a function of time, and simultaneously samples were extracted at different time intervals which shows that the single atoms agglomerate to clusters and then to nanoparticles. The sample extracted after 6 min, immediately after the induction period when the catalytic reaction started to show the small cluster size of 4-13 gold atoms as presented in Fig. 8(b). After 120 min of reaction [Fig. 8(c)], the activity of the catalyst stopped. At this point, gold had aggregated into nanoparticles of diameter ≥2 nm. The population distribution as a function of time is shown in Fig. 8(d). For an alternative verification of the performance, isolated gold clusters with 3, 5–7, and 7–10 atoms were prepared separately and deposited on functionalized MWCNTs following the procedure described in Ref. 96 and the results are shown in Fig. 8(e). From these results, we can conclude that the most-active gold species for the oxidation of thiophenol to benzene disulfide correspond to small gold clusters that contain between five and ten atoms.⁹⁶ DFT calculations reveal that thiophenol adsorbs on all these clusters to form a stable Au–S bond and transfer charge from the S lone pair to metal LUMO. O₂ interaction with the clusters results in a noticeable activation of the O–O bond as shown in Fig. 8(f). Coadsorption of thiophenol and O₂ is always energetically favourable, and in some cases a synergistic effect that leads to a higher degree of O₂ activation is observed. The reaction mechanism for Au₅ is shown in Fig. 8(g) which proceeds via coadsorption of thiophenol and O₂ on two neighbouring Au atoms, leading to the formation of H₂O₂ which eventually decomposes to two OH groups resulting in the formation of H₂O. After H₂O₂ desorption, the two thiolate fragments that remain adsorbed on the Au₅ cluster [Fig. 8(g)] are nearly neutral as shown in Fig. 8(h). Therefore, there is no electrostatic repulsion between the two S atoms, and the S–S bond can be formed that also involves an important rearrangement of the gold-cluster geometry. Au atoms and Au_n clusters differ in the ability of the latter to transfer electrons from thiolate fragments to O₂, which makes the formation of an S–S bond possible, as happens with sulfhydryl oxidase enzymes. For larger particles, a drop in the activity of disulphide formation is noticed as along with poisoning of the nanoparticles with thiolate groups. This is due to the lower concentration of active Au atoms accessible to reactants as particles grow, and by the strong Au–S interaction leading to the formation of very stable motifs and passivation of the gold nanoparticle surface [Fig. 8(i)].^97,98 However, the catalytic activity of size-selected unsupported thiol-protected metal clusters has been more closely investigated recently and shown in specific examples to be significant,⁹⁹ and thus the Au–S interactions may not always be detrimental to chemical activity.

The model cluster systems that we have summarized in this perspective are ideal for a fundamental understanding of the catalytic properties, as well as well suited to identify new materials for chemical and electrochemical transformations, while they can also serve to validate the function of new materials designed by computational screening. However, most of the typical fabrication methods discussed so far that utilize physical gas phase production is designed for model/laboratory scale studies. These methods were not conceived with the goal to produce massive amounts of catalysts for industrial applications. However, the very promising, and in several cases unique, properties of these cluster-based materials call for a scale up of their production. This has become an important endeavour, with some efforts utilizing the scale up of the very physical methods of size-selected cluster production through technical modifications,^100–105 and, very importantly also by employing alternative wet-chemical, solution-based methods to fabricate sizeable amounts of size-selected catalysts with very narrow size distributions,^96,106–108 also combining wet-chemical methods of a precursor that is then deposited and thermally decomposed to produce size-selected cluster catalysts supported on ionic surfaces such as zeolites¹⁰⁹ or anchored to carbon nanotubes.¹¹⁰ However, in some cases, the ligand effect might play a role in the catalyst performance.^111,112 Other advanced synthesis methods have been reported for ligand-free preparation of Ag,¹¹³ Au,⁹⁶ and Ir¹¹⁴ clusters, for example.

Recent development of new generations of cluster beam sources based on magnetron sputtering has successfully generated size-selected clusters of mass in the order of grams per hour.¹⁰⁰ This is several orders of magnitude higher than that produced in conventional cluster production sources that have been summarized above. The clusters are supported on (graphite) powders suitable for realistic test tube reaction conditions.¹¹⁵ The catalyst powders were synthesized by cluster beam deposition onto the planar surface followed by microdicing. The catalysts are highly stable and sinter resistant. Performance of cluster-based catalysts has been compared with catalysts developed using the conventional “wet chemistry” technique. The catalysts have been tested for hydrogenation reactions which demonstrated great performance and in some cases outperform those synthesised following conventional methods.¹⁰⁰ There have also been other developed techniques like storing the clusters within a rare gas matrix.¹⁰¹ Some other recent developments for cluster synthesis using a magnetron sputtering source include high-power impulse magnetron sputtering¹⁰⁴ or optimized design for generation of bimetallic and trimetallic alloy clusters.¹⁰² 

In the present review, we have tried to illustrate, via selected but representative showcase examples and accompanying discussion, the recent developments in the synthesis, experimental and theoretical characterization, and catalytic performance of size-selected subnanometer supported cluster catalysts.

A few key points can be derived from this overview.

First and foremost, the fact that these systems can exhibit an unconventional and sometimes extremely high catalytic activity and selectivity, so that they may rank among the most active catalysts reported to date and at the same time may be also viewed as models of the active site.

Second, the non-trivial dependence of the chemical properties of these systems on (i) their size/composition (such that adding or removing a single metal atom or replacing a single metal element with a dopant can dramatically alter their functions), (ii) the substrate on which they are deposited, and (iii) the reaction conditions (concept of ligand/cluster/support catalytic complex).

These two features make size-selected subnanometer clusters very promising and appealing systems in the search of novel, more active, and selective catalysts as required by present environmental and societal requirements, e.g., functioning under milder conditions of low temperature and pressure and/or minimizing the use of precious metals, as compared to bulk or nanosized catalysts of identical materials.

At the same time, because of these two same features, it is clear that some form of advanced characterization is necessary to shed light on reaction mechanisms and intermediates and thus to unveil and understand in depth the sometimes exotic chemistry of these systems, so as to fully realize their catalytic potential. Hence the important role played in recent advances by progress in both in situ experimental characterization techniques and in theory and computational modeling, for the unique capability of the latter to quantitatively test hypotheses of catalytic pathways and the associated sequence of configurations and transitions among them.

These combined and synergic experimental/theoretical studies clearly singled out the oxidation state of the metal elements (also depending on the charge transfer to/from the substrate and the adsorbates¹¹⁶) and the combined effect of the substrate and the ligand shell surrounding the cluster under realistic conditions, as two key factors determining catalytic behaviour. The ligand shell in fact operates together with cluster/substrate interactions to stabilize such small sub-nanometer and under-coordinated cluster species against disaggregation and/or diffusion over the substrate, leading to sintering and possible catalyst deactivation. In this respect, a crucial ingredient appears to be the amorphous character of the substrate, a fascinating subject but also extremely challenging in view of the issues it poses to both experimental characterization and theoretical (stochastic sampling) modeling.

The preceding discussion of the ligand effects naturally provides paths of future developments and progress as well. The importance of ligand high-coverage phases and of determining the ligand/cluster/support catalytic complex formed in situ under the environmental conditions of the specific catalytic reaction is one avenue which seems particularly fruitful and worthwhile elaborating. For example, the role played by catalyst preparation in real-world catalytic systems is well known, to a point that they often appear as “black magic” protocols.¹¹⁷ In contrast to heterogeneous catalysis by larger nanoparticles where these techniques have been thoroughly investigated, such opportunities have not been emulated so far and are basically novel to the field of sub-nanometer catalysts. To quote one example, the use of a catalytic complex produced under the conditions of a given catalytic reaction as the catalyst in a different catalytic process has only been hypothesized or suggested theoretically but—to the best of our knowledge—never explored experimentally. Analogously, the introduction of additives and promoters into the reaction medium has not been explored so far, although it offers exciting perspectives especially once the details of the reaction mechanism have been appropriately revealed.

Another avenue which seems very promising is that of heterogeneous photo-catalysis. There is not so much knowledge on the optical properties of sub-nanometer supported metal clusters, see, e.g., Ref. 118, but it can be hypothesized, also by analogy with analogous size-selected systems,¹¹⁹ that they can give rise to a rich photochemical behavior. Other unexplored avenue holding great potential is bond-selective electrochemical synthesis yielding gas, liquid, and solid phase products.

This material is based upon work supported by the U.S. Department of Energy, Office of Science, Office of Basic Energy Sciences Materials Science and Engineering (A.H., L.A.C, and S.V.) under Contract No. DE-AC-02-06CH11357. A.F. gratefully acknowledges support from the ERC-AG SEPON project, Italian CNR, and the CINECA Supercomputing Centre within the ISCRA Programme, and the use of the Center for Nanoscale Materials, an Office of Science user facility, was supported by the U.S. Department of Energy, Office of Science, Office of Basic Energy Sciences, under Contract No. DE-AC02-06CH11357.